Vref+ as Output on MSP430FR5739

Many thanks Katie for your indications.

The tests are based on the Experimenter Board MSP- EXP430FR5739.

The NTC implementation on the Experimenter Board also uses a GPIO pin P2.7 to source the NTC device (see below detail from circuit diagram, User Guide slau343b.pdf).
NTC has 100k resistance and R35 has 470k resistance. This gives at room temperature a sum of 570k Ohm.

In my NTC circuit I also use an NTC with 100k (at 25°C) but a smaller resistor (in place of R35) of 100k. So the sum is about 200k Ohm at room temperature (now 22°C). Using a power source with 3.6 V you would have a current of 0.018 mA which can easily be supplied from a GPIO pin. So the problem should not come from this side.

On the board we have a regulated voltage (from TPS77301DGK) of Vcc = 3.67 V (measured).

Measurement setup for my configuration:
Power source from pin P2.5:                      Vcc       = 3.670 V            (Variation: +/- 0.010 V)
Measuring NTC signal at pin P1.3:             Vsignal  = 1.760 V            (Variation: +/- 0.003 V)
Vref is set to:                                                  Vref+     = Vcc = 3.670 V
                                                                         Vref -     = GND   (I guess this is equivalent to AVss)
3.67 V / 1023 bit    results in                       3.587 mV/bit

Further analysis:
1. To check if the ADC is working well I have first connected the output pin P2.5 (3.670 V) to the analog input pin P1.5 to see the max. conversion result (full range signal at input pin).
   ADC10MEMO gives a value of 1022    (Variation: +/- 1;  during 50 measurements)
This variation range of 2 units seems to me (at first sight) stable and OK.

2. Connect the output pin P2.5 to power the NTC circuit with 3.67 V. Now I can measure an NTC signal of 1.760 V as input for pin P1.3. From that value I would expect a conversion result of                1.76/3.67 * 1023             =  491 units.
But reading the ADC10MEMO register I get values varying from 454 to 547 (during 50 measurements).
This is a variation range of 93 units (+56 / -37  with an equivalent of +200mV / - 133 mV). Compared to the range from case 1., it is about 50 times worse.

3. When I drive the NTC circuit directly from Vcc on the board I get the same bad results as in case 2.

4. When I look at the ADC10MEMO register running the “User experience” program for NTC temperature measurement (coming with the Experimenter Board), I see stable conversion results.

5. I also connected the NTC circuit in the same way to a Launchpad with MSP430G2553. There I got the same unsatisfying results as described in point 2..

Conclusions:
A. Vcc seems to be stable enough (see 1.: very small variation in ADC result when P2.5 (Full span of 3.67 V) directly connected to P1.3 (Analog In)).
B. Both, powering the NTC voltage divider circuit from GPIO pin or from Vcc shall be possible, because the current over the NTC circuit is very low (<< 0.1 mA).

So what is still the problem when Vcc is stable enough and GPIO can power the low current of NTC circuit ???
I have made a number of changes to ADC setting (e.g. sample + hold time variations, changing channels, changing GPIOs), all without positive effects.

Solution:
After hours of searching and reading, I found this posting from Jason Sachs:

     www.embeddedrelated.com/.../91.php&nbsp;

Then I applied an R-C-Filter between the analog input pin and the signal from NTC.
Incredible, that was it. The conversion result is now stable.

Don’t know if TI has an application note describing the technique of filtering an analog signal before ADC. If you have please let me know. This could avoid a lot of trouble, perhaps for other users too.

Helmut

Are you using a minimal sampling time of 9uS?

Ref. to:

An accurate reference voltage is needed only if you want to measure voltage in absolute terms, which is not the case for a NTC voltage divider.

Yes, this is good to remember.

Ref to:

The reference voltage source on your own board may react badly when the ADC takes samples. That is probably why you need to add RC in-between to soften the blow of ADC taking sample. 

But why do I get stable results when the AVcc is fed directly to the analog Input (disconnected NTC circuit) ? I would expect the same variation in the conversion result.  Because it is the same source and the same ADC configuration. This leads to the conclusion that the NTC circuit is the limiting part.

fritz01h said:

But why do I get stable results when the AVcc is fed directly to the analog Input (disconnected NTC circuit) ? I would expect the same variation in the conversion result.  Because it is the same source and the same ADC configuration. This leads to the conclusion that the NTC circuit is the limiting part.

Difference in series resistance = difference in ADC capacitor charging time!

Then you are using a wrong Sampling time, but your sampling time will not be the only error in your setup.

For the Sampling time read the ‘User’s Guide’ section “16.2.5.3 Sample Timing Considerations” additional data you can find in the ‘Data Sheet’ “5.28 10-Bit ADC, Power Supply and Input Range Conditions” and further.

The ADC connects to, and takes a snapshot (sample) of the to be measured voltage and stores this into a capacitor which, after disconnecting from the source/input, will be evaluated (converted) to a digital value.

But as any capacitor (or maybe you understand better ‘battery’) the charge takes time, normally expressed in T=RxC.

R is not a single resistor and C is not a single capacitor, in chapter 16.2.5.3 of the User’s Guide you see the equivalent of all of them.

Some of the values are specific to the device which you can read in the Data Sheet other are depending on board/wiring capacitance and series resistor value (470K + 100K).

From these parameters you need to calculate the minimum sample time needed for a good charge of the ADC capacitor to a representative voltage value. The time RxC gives you the time to a charge of ~63%, 5xRxC is near 100%, but for a more precise conversion result you need to take 8 (Tau) times (8xRxC).

Maybe you now understand that when connecting your ADC input directly to your supply voltage the series resistance is very low and the capacitor charges fast, and just a short sampling time is needed. But in your NTC setup the series resistance is much larger and so the time needed to charge the capacitor much larger and a short sampling time is not sufficient to fully charge the capacitor.

Buffering the NTC value with an extra capacitor connected to the ADC input pin will (partly) eliminate the NTC circuit resistance and a shorter sample time will work, but still needs to be calculated.

At the moment I’m too far away from being able to do the required calculations. But I understand that I have to increase the sample time to get the complete signal captured. So first l would like to continue configuring the ADC unit for longer sample periods.

I have set the ADC10SHSx to max. value. The ADC conversion value is still changing too much.
Then I set the ADC10DIVx to the highest value to see what happens. Yes, the value variation has now decreased. But the span is still not acceptable.
Ok there is another chance to increase the sample time by setting ADCPDIVx to the highest value. Unfortunately also this gives not the desired result. The conversion value is not stable.

I do not understand why the NTC circuit on the Experimenter Board works fine. It is also an NTC with 100 k resistance (see my earlier descriptions).

Perhaps the best solution would be adding a unity-gain amplifier between NTC signal and analog input pin of the MCU.

Regarding the required calculations, do you know some description that I could study to learn this?
I also asked TI for an application note. No answer till now.

Many thanks for your effort.

Helmut

Katie,

I want to use the external reference as an offset to an external amplifier with a single supply. But you said the internal reference cannot be used to drive another device? So what does this mean in Section 4.6?

And this register description: